Implementing the routing switch

Implementing efficient routing on a network can increase performance whilst saving money

Implementing efficient routing on a network can increase performance whilst saving money.

Today's applications are consuming bandwidth at ever-growing rates. Although this statement originally referred to memory, it seems to fit today's networks equally well. New applications and speedier processors are placing unprecedented demands on the network ( and we have only seen the beginning. Voice, video, multicast, push technology ( these and other bandwidth-hungry applications are on the rise. Together they have the potential to rewrite all the rules.

A routing switch represents a new class of switch. It offers the Layer 3 forwarding historically found in routers with the ultra high-speed performance and low price point normally associated with traditional Layer 2 switches.

Designed and optimised from inception using hardware-based ASICs, routing switches feature wire-speed Layer 2 and Layer 3 forwarding, as well as flexible, advanced features including Gigabit Ethernet, Layer 4 prioritisation, IP multicast, VLANs and QoS.

Routing switches can be used in numerous applications including:

1) Providing high-speed Layer 2 aggregation (with advanced features at a traditional Layer 2 switch price point)

2) Offloading IP traffic from backbone routers (making the traditional router more efficient for legacy protocols, firewalling, and WAN connectivity)

3) Building server farms and servicing power workgroups

And because routing switches accomplish all of this using standards-based protocols, they can be easily integrated into existing infrastructures.

The need for routing switches

New users and bandwidth-hungry applications like the Web are causing exponential growth in network traffic. IT managers and engineers have spent the last few years turning heterogeneous workgroup networks into robust, connected, enterprise-wide resources. Nevertheless, today's applications and users demand still more. The resulting challenge is to migrate the enterprise network to a high-performance IP infrastructure, a creation sometimes referred to as an intranet. If there were any doubt about the need to meet this challenge, the advent of the Web has eliminated it. Web technology creates online-information addicts, luring users ever further beyond their local workgroups. Enablers like IP Multicast and ReSerVation Protocol (RSVP) offer application developers new tools for bringing rich content to the desktop. Though local resources were once the norm, these new applications are anything but local.

Today's networks must be able to handle anywhere-to-anywhere traffic. This new reality spells the death of the 80/20 rule, a rule that guided network design for over a decade.

IT managers, confronted with increasing loads and unpredictable traffic patterns, must develop a plan to ensure the performance of mission-critical applications. Because inter-subnet routers have become bottlenecks, existing networks are hard-pressed to process unpredictable traffic efficiently.

The addition of high-performance switching can increase subnet size and reduce the number of router ports required. However, there are limits to the size of switched networks, whether physical or logical (VLANs). It is clear that both switching and routing are still required elements of a good campus-network design. Unfortunately, there is a large and growing performance gap between existing routers and switches. Traditionally, network designers have built their networks around router hops whenever possible. But today's unpredictable traffic flows can undermine the assumptions of network designers. And they're liable to do so at any given time. Clearly, network designers need a solution that can adapt to increasing traffic demands, unpredictable traffic flows, and the priority of mission-critical applications.

Proposals for performance improvement

As IT managers confront the changing requirements outlined above, the industry is awash with proposals. In all cases, switching is a central theme, but proposals differ when it comes to routing. Though each can be very complex and require much time to evaluate, the proposals can be broken down into two fundamental approaches:

1) Avoid routing

Most industry efforts over the past few years have centered on the first approach: avoid routing. These schemes either "route once, switch many" or try to avoid routing altogether by implementing a completely flat network. New protocols and technologies like IP Switching from Ipsilon, FastIP from 3Com, and SecureFast Virtual Networking from Cabletron is evidence of the considerable time and effort expended. However, the "no free lunch" rule applies to all approaches that avoid routing.

Standard Layer 2 switches have limitations that preclude building large flat networks. Most importantly, standard switches cannot contain broadcasts without breaking the network into pieces, either physically or logically (VLANs). Hence, any scheme to avoid routing must utilise modified Layer 2 switches. Consequently, any "route once, switch many" or flat network strategy cannot achieve large scalable networks without introducing new functionality into the switch. The result is that all of these schemes introduce new protocols and/or network components to the network, are proprietary or have very limited support from a handful of vendors, require substantial changes to the network to achieve the desired result.

2) Increase router performance

All of this complexity and difficulty has led many in the industry to ask the obvious questions: why don't we fix the problem rather than avoid it? Can't we build routers that achieve switching speeds? In the past, the answers have been "because we can't" and "no" respectively. But much has changed. It is now feasible to attack the problem head-on.

Routing switches

A new class of switch is needed to avoid the tradeoffs highlighted above - added complexity versus inadequate performance. By breaking from the processor implementation of traditional routers, routing switches set new standards for cost and performance. The concepts are simple: apply switching techniques to only those protocols that require optimised routing performance, and fully integrate high performance routing into the switch fabric.

If implemented correctly, this solution enables routers to function at switching speeds. When building a switched network, you can solve two problems by deploying routing switches rather than Layer 2 switches:

( You gain the capacity to aggregate large switched networks, and

( You gain the ability to flexibly offload IP routing from your current backbone routers

That means much higher throughput. In effect, routing switches allow you to build a high-performance intranet within your current network. A routing switch can be initially deployed as a very capable Layer 2 switch. But unlike a typical Layer 2 switch, a routing switch can provide IP routing. In each case, switching and IP routing are performed at wire speed with microsecond latency.

Benefits of routing switches

Routing switches eliminate the performance penalty previously associated with Layer 3 traffic. Consequently, they greatly simplify network design and yield enormous benefits to IT managers.

"No Penalty" routing makes network design simpler

It is common practice when designing networks to avoid router hops and their attendant latency. For example, network designers often avoid router hops by including local servers on the same subnet as primary clients. However, when you locate the server in the data center rather than in the workgroup, you must add extra switch and riser links. With routing switches you don't need to avoid router boundaries, so you eliminate the concern altogether.

Quality of Service (QoS) supports advanced applications

For some applications, raw bandwidth is not enough. Multimedia streams require consistent latencies even when a bursty transfer occurs while a stream is traversing the net. A well-designed routing switch provides priority queuing. This enables it to provide consistent latency across entire intranets without resorting to cut-through routing schemes. Whereas current processor-based routers can offer only software-type latencies for protocols such as RSVP, routing switches provide the same functionality at switching latencies.

For example, you could deploy IP Multicast, or even a video server that can play Moving Picture Experts Group (MPEG)-1 video, across an entire campus or corporate network. And you could do so in a fashion that's simple and robust. By fully integrating routing and switching technology, you open up possibilities previously considered the domain of circuit-switched technologies, such as ATM.

Because routing switches don't introduce new protocols, they integrate seamlessly within the existing network infrastructure. Remarkably flexible, they can operate initially as high-performance Layer 2 switches. Then, once configured for routing, they can increase IP performance and reduce the burden on the existing network router, extending its useful life.

Almost all switches and routers slow dramatically when their advanced features are turned on. For instance, turn on policy-based VLANs and the performance of some switches can drop by over 50 per cent. Likewise, routers that implement priorities suffer huge performance deficits when those features are activated.

The reason is straightforward: typical switches and routers rely on a CPU to provide advanced features. CPUs are a shared resource; the more processes that run on them, the slower they execute each one. Well-designed routing switches avoid this undesirable characteristic through superior architectures.

Another benefit of routing switches is integrated management. Out of the box, routing switches operate as Layer 2 switches. You can configure VLANs, routing, or IP Multicast from one console. You can even set up two VLANs and route between them with ease. Since routing and switching are combined, inter-VLAN routing does not consume front panel ports. A Media Access Control (MAC) address for the router "appears" on the VLAN, completely configured within the switch.

( Nortel Networks 1999

Compiled by Mike Burkitt

Read more on Voice networking and VoIP